A little background: Me and two friends have to come up with an AI project by the end of the year. Our marks will pretty much be determined 100% by this project. The class is made up of mostly grad students, theres only about 10 undergrads. So... it'd be really nice to get a good mark on this project.

We were told to make it ambitious so our first set of ideas are probably not do-able, but I thought I'd get AI-depot's opinions on a few of them.

1.The first idea is to go get training data from www.hotornot.com. We'd filter through around 10 ,000 pictures and try to narrow the training data down to single head shots. We'd then train our neural net to predict wether a person would be hot or not. This would be a cool project from a philosophy point of view, can human attraction be predicted by a computer?

2.Another idea was to take a satellite feed and weather data from various sources and see how long we could predict the weather for.

Both of these seem pretty hard considering we haven't done much image analysis before.

3. A third idea was to make a game AI for quake 2 or some other open source game. Kind of like reaper bot or oak bot or something but using the theoretical AI we've been taught at school, and using much more powerful modern day cpus. I doubt this project would be too hard.

4. Predict the stock market in some way. I think this would probably be the easiest of all the projects. You could implement the Cox-Ross-Rubinstein model, find the expected value for every stock out of about 5000, and us that to predict if the overall market would go up or down tomorrow. I've already got the training data for this one, 5000 stocks over the last 4 years or so with all the data we'd need. I'd also be interested in data mining algs to find the best 10 stocks to invest in etc, but the overall market would probably be the easiest.

Anyway, I've put theses projects in the order from hardest to easiest according to what I know. Anyone have any information on how hard 1+2 would actually be? I would love to be able to do number 1 but it really does seem like an impossibly hard project.

All of these seem pretty ambitious to me. I personally like the first idea, but don't know how feasible it is. What tools (software, programming skills)do you have at your disposal?

-Predictor
http://will.dwinnell.com

Ya they are all pretty difficult. I'm not really sure if we'd be able to do them or not. We have 3 coders at the moment, all 4th year computing students who are taking their second class on AI. We program mostly in java/c/c++ but know other langauges. We have access to all the development tools there as well as matlab and some AI specific software, although we've never had to use the AI software. We've just been using matlab so far in this course and used java the course before for robo-cup, wumpa world etc.

The main challenge I think for #1 would be the image processing, anyone know if there are any easy was to abstract, or address that problem?

Even without the Image Processing Toolbox, MATLAB is great for image processing (bitmap images are arrays, which is what MATLAB does best). You might give this project a try.

-Predictor
http://will.dwinnell.com